Skip to content

[AMDGPU] Add gfx1250 wmma_scale[16]_f32_32x16x128_f4 instructions #152194

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 6 commits into from
Aug 5, 2025
Merged

[AMDGPU] Add gfx1250 wmma_scale[16]_f32_32x16x128_f4 instructions #152194

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
27300f7
[AMDGPU] Add gfx1250 wmma_scale[16]_f32_32x16x128_f4 instructions
rampitec Aug 5, 2025
98e9fb4
Merge branch 'main' into users/rampitec/08-05-_amdgpu_add_gfx1250_wmm…
rampitec Aug 5, 2025
b86e38e
Merge branch 'main' into users/rampitec/08-05-_amdgpu_add_gfx1250_wmm…
rampitec Aug 5, 2025
9839ae6
Merge branch 'main' into users/rampitec/08-05-_amdgpu_add_gfx1250_wmm…
rampitec Aug 5, 2025
a3abcd2
Merge branch 'main' into users/rampitec/08-05-_amdgpu_add_gfx1250_wmm…
rampitec Aug 5, 2025
3287cf4
Merge branch 'main' into users/rampitec/08-05-_amdgpu_add_gfx1250_wmm…
rampitec Aug 5, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions clang/include/clang/Basic/BuiltinsAMDGPU.def
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -805,6 +805,8 @@ TARGET_BUILTIN(__builtin_amdgcn_wmma_scale16_f32_16x16x128_f8f6f4, "V8fIiV16iIiV
TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x32_f16, "V8fIbV16hIbV16hIsV8fIbIb", "nc", "gfx1250-insts,wavefrontsize32")
TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x32_f16, "V8hIbV16hIbV16hIsV8hIbIb", "nc", "gfx1250-insts,wavefrontsize32")
TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_32x16x128_f4, "V16fV16iV8iIsV16f", "nc", "gfx1250-insts,wavefrontsize32")
TARGET_BUILTIN(__builtin_amdgcn_wmma_scale_f32_32x16x128_f4, "V16fV16iV8iIsV16fIiIiiIiIiiIbIb", "nc", "gfx1250-insts,wavefrontsize32")
TARGET_BUILTIN(__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4, "V16fV16iV8iIsV16fIiIiLiIiIiLiIbIb", "nc", "gfx1250-insts,wavefrontsize32")
TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x64_bf16, "V8fIbV16yIbV32yV8fiIbIb", "nc", "gfx1250-insts,wavefrontsize32")
TARGET_BUILTIN(__builtin_amdgcn_swmmac_bf16_16x16x64_bf16, "V8yIbV16yIbV32yV8yiIbIb", "nc", "gfx1250-insts,wavefrontsize32")
TARGET_BUILTIN(__builtin_amdgcn_swmmac_bf16f32_16x16x64_bf16, "V8fIbV16yIbV32yV8fiIbIb", "nc", "gfx1250-insts,wavefrontsize32")
Expand Down
10 changes: 10 additions & 0 deletions clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -894,6 +894,8 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
case AMDGPU::BI__builtin_amdgcn_wmma_f32_32x16x128_f4:
case AMDGPU::BI__builtin_amdgcn_wmma_scale_f32_16x16x128_f8f6f4:
case AMDGPU::BI__builtin_amdgcn_wmma_scale16_f32_16x16x128_f8f6f4:
case AMDGPU::BI__builtin_amdgcn_wmma_scale_f32_32x16x128_f4:
case AMDGPU::BI__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4:
case AMDGPU::BI__builtin_amdgcn_swmmac_f32_16x16x64_f16:
case AMDGPU::BI__builtin_amdgcn_swmmac_f32_16x16x64_bf16:
case AMDGPU::BI__builtin_amdgcn_swmmac_f16_16x16x64_f16:
Expand Down Expand Up @@ -1172,6 +1174,14 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
ArgsForMatchingMatrixTypes = {3, 0, 1};
BuiltinWMMAOp = Intrinsic::amdgcn_wmma_f32_32x16x128_f4;
break;
case AMDGPU::BI__builtin_amdgcn_wmma_scale_f32_32x16x128_f4:
ArgsForMatchingMatrixTypes = {3, 0, 1};
BuiltinWMMAOp = Intrinsic::amdgcn_wmma_scale_f32_32x16x128_f4;
break;
case AMDGPU::BI__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4:
ArgsForMatchingMatrixTypes = {3, 0, 1};
BuiltinWMMAOp = Intrinsic::amdgcn_wmma_scale16_f32_32x16x128_f4;
break;
case AMDGPU::BI__builtin_amdgcn_swmmac_f32_16x16x64_f16:
ArgsForMatchingMatrixTypes = {4, 1, 3, 5};
BuiltinWMMAOp = Intrinsic::amdgcn_swmmac_f32_16x16x64_f16;
Expand Down
22 changes: 22 additions & 0 deletions clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250-wmma-w32.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -314,6 +314,28 @@ void test_amdgcn_wmma_f32_wmma_f32_32x16x128_f4(global v16f* out, v16i a, v8i b,
*out = __builtin_amdgcn_wmma_f32_32x16x128_f4(a, b, 0, c);
}

// CHECK-GFX1250-LABEL: @test_amdgcn_wmma_scale_f32_32x16x128_f4(
// CHECK-GFX1250-NEXT: entry:
// CHECK-GFX1250-NEXT: [[TMP0:%.*]] = tail call <16 x float> @llvm.amdgcn.wmma.scale.f32.32x16x128.f4.v16f32.v16i32.v8i32(<16 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], i16 0, <16 x float> [[C:%.*]], i32 1, i32 2, i32 [[SCALE_SRC0:%.*]], i32 2, i32 1, i32 [[SCALE_SRC1:%.*]], i1 false, i1 true)
// CHECK-GFX1250-NEXT: store <16 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 64, !tbaa [[TBAA4]]
// CHECK-GFX1250-NEXT: ret void
//
void test_amdgcn_wmma_scale_f32_32x16x128_f4(global v16f* out, v16i a, v8i b, v16f c, int scale_src0, int scale_src1)
{
*out = __builtin_amdgcn_wmma_scale_f32_32x16x128_f4(a, b, 0, c, 1, 2, scale_src0, 2, 1, scale_src1, 0, 1);
}

// CHECK-GFX1250-LABEL: @test_amdgcn_wmma_scale16_f32_32x16x128_f4(
// CHECK-GFX1250-NEXT: entry:
// CHECK-GFX1250-NEXT: [[TMP0:%.*]] = tail call <16 x float> @llvm.amdgcn.wmma.scale16.f32.32x16x128.f4.v16f32.v16i32.v8i32(<16 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], i16 0, <16 x float> [[C:%.*]], i32 1, i32 2, i64 [[SCALE_SRC0:%.*]], i32 2, i32 1, i64 [[SCALE_SRC1:%.*]], i1 false, i1 true)
// CHECK-GFX1250-NEXT: store <16 x float> [[TMP0]], ptr addrspace(1) [[OUT:%.*]], align 64, !tbaa [[TBAA4]]
// CHECK-GFX1250-NEXT: ret void
//
void test_amdgcn_wmma_scale16_f32_32x16x128_f4(global v16f* out, v16i a, v8i b, v16f c, long scale_src0, long scale_src1)
{
*out = __builtin_amdgcn_wmma_scale16_f32_32x16x128_f4(a, b, 0, c, 1, 2, scale_src0, 2, 1, scale_src1, 0, 1);
}

// CHECK-GFX1250-LABEL: @test_amdgcn_swmmac_f32_16x16x64_bf16(
// CHECK-GFX1250-NEXT: entry:
// CHECK-GFX1250-NEXT: [[TMP0:%.*]] = tail call <8 x float> @llvm.amdgcn.swmmac.f32.16x16x64.bf16.v8f32.v16bf16.v32bf16.i32(i1 false, <16 x bfloat> [[A:%.*]], i1 false, <32 x bfloat> [[B:%.*]], <8 x float> [[C:%.*]], i32 [[INDEX:%.*]], i1 false, i1 true)
Expand Down
26 changes: 26 additions & 0 deletions clang/test/SemaOpenCL/builtins-amdgcn-error-gfx1250-wmma-w32-param.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -230,6 +230,32 @@ void test_amdgcn_wmma_f32_wmma_f32_32x16x128_f4(global v16f* out, v16i a, v8i b,
*out = __builtin_amdgcn_wmma_f32_32x16x128_f4(a, b, mod, c); // expected-error {{'__builtin_amdgcn_wmma_f32_32x16x128_f4' must be a constant integer}}
}

void test_amdgcn_wmma_scale_f32_32x16x128_f4(global v16f* out, v16i a, v8i b, v16f c, int mod, int scale_src0, int scale_src1, bool reuse)
{
*out = __builtin_amdgcn_wmma_scale_f32_32x16x128_f4(a, b, mod, c, 1, 0, scale_src0, 2, 0, scale_src1, 1, 0); // expected-error {{'__builtin_amdgcn_wmma_scale_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale_f32_32x16x128_f4(a, b, 0, c, mod, 0, scale_src0, 2, 0, scale_src1, 0, 0); // expected-error {{'__builtin_amdgcn_wmma_scale_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, mod, 0, scale_src1, 0, 0); // expected-error {{'__builtin_amdgcn_wmma_scale_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, 2, 0, scale_src1, reuse, 0); // expected-error {{'__builtin_amdgcn_wmma_scale_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, 2, 0, scale_src1, 0, reuse); // expected-error {{'__builtin_amdgcn_wmma_scale_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale_f32_32x16x128_f4(a, b, 0, c, 1, mod, scale_src0, 2, 0, scale_src1, 1, 0); // expected-error {{'__builtin_amdgcn_wmma_scale_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, 2, mod, scale_src1, 1, 0); // expected-error {{'__builtin_amdgcn_wmma_scale_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, 2, 0, scale_src1, mod, 0); // expected-error {{'__builtin_amdgcn_wmma_scale_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, 2, 0, scale_src1, 1, mod); // expected-error {{'__builtin_amdgcn_wmma_scale_f32_32x16x128_f4' must be a constant integer}}
}

void test_amdgcn_wmma_scale16_f32_32x16x128_f4(global v16f* out, v16i a, v8i b, v16f c, int mod, long scale_src0, long scale_src1, bool reuse)
{
*out = __builtin_amdgcn_wmma_scale16_f32_32x16x128_f4(a, b, mod, c, 1, 0, scale_src0, 2, 0, scale_src1, 1, 0); // expected-error {{'__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale16_f32_32x16x128_f4(a, b, 0, c, mod, 0, scale_src0, 2, 0, scale_src1, 0, 0); // expected-error {{'__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale16_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, mod, 0, scale_src1, 0, 0); // expected-error {{'__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale16_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, 2, 0, scale_src1, reuse, 0); // expected-error {{'__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale16_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, 2, 0, scale_src1, 0, reuse); // expected-error {{'__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale16_f32_32x16x128_f4(a, b, 0, c, 1, mod, scale_src0, 2, 0, scale_src1, 1, 0); // expected-error {{'__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale16_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, 2, mod, scale_src1, 1, 0); // expected-error {{'__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale16_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, 2, 0, scale_src1, mod, 0); // expected-error {{'__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4' must be a constant integer}}
*out = __builtin_amdgcn_wmma_scale16_f32_32x16x128_f4(a, b, 0, c, 1, 0, scale_src0, 2, 0, scale_src1, 1, mod); // expected-error {{'__builtin_amdgcn_wmma_scale16_f32_32x16x128_f4' must be a constant integer}}
}

void test_amdgcn_swmmac_f32_16x16x64_bf16(global v8f* out, v16bf16 a, v32bf16 b, v8f c, int index, int mod)
{
*out = __builtin_amdgcn_swmmac_f32_16x16x64_bf16(mod, a, 0, b, c, index, false, false); // expected-error {{'__builtin_amdgcn_swmmac_f32_16x16x64_bf16' must be a constant integer}}
Expand Down
24 changes: 24 additions & 0 deletions llvm/include/llvm/IR/IntrinsicsAMDGPU.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3956,6 +3956,28 @@ class AMDGPUWmmaScaleIntrinsicModsC<LLVMType scale_ty> :
IntrWillReturn, IntrNoCallback, IntrNoFree]
>;

class AMDGPUWmmaScaleF4IntrinsicModsC<LLVMType scale_ty> :
Intrinsic<
[llvm_anyfloat_ty], // %D
[
llvm_anyint_ty, // %A
llvm_anyint_ty, // %B
llvm_i16_ty, // %C_mod: 0 - none, 1 - neg, 2 - abs, 3 - neg(abs)
LLVMMatchType<0>, // %C
llvm_i32_ty, // matrix_a_scale
llvm_i32_ty, // matrix_a_scale_fmt
scale_ty, // matrix a scale exponential
llvm_i32_ty, // matrix_b_scale
llvm_i32_ty, // matrix_b_scale_fmt
scale_ty, // matrix b scale exponential
llvm_i1_ty, // matrix_a_reuse
llvm_i1_ty, // matrix_b_reuse
],
[IntrNoMem, IntrConvergent, ImmArg<ArgIndex<2>>, ImmArg<ArgIndex<4>>, ImmArg<ArgIndex<5>>, ImmArg<ArgIndex<7>>,
ImmArg<ArgIndex<8>>, ImmArg<ArgIndex<10>>, ImmArg<ArgIndex<11>>,
IntrWillReturn, IntrNoCallback, IntrNoFree]
>;

defset list<Intrinsic> AMDGPUWMMAIntrinsicsGFX1250 = {
def int_amdgcn_wmma_f32_16x16x4_f32 : AMDGPUWmmaIntrinsicModsAllReuse<llvm_anyfloat_ty, llvm_anyfloat_ty>;
def int_amdgcn_wmma_f32_16x16x32_bf16 : AMDGPUWmmaIntrinsicModsAllReuse<llvm_anyfloat_ty, llvm_anyfloat_ty>;
Expand Down Expand Up @@ -3984,6 +4006,8 @@ def int_amdgcn_wmma_f32_16x16x128_f8f6f4 : AMDGPUWmmaIntrinsicModsC_MatrixFMT;
def int_amdgcn_wmma_scale_f32_16x16x128_f8f6f4 : AMDGPUWmmaScaleIntrinsicModsC<llvm_i32_ty>;
def int_amdgcn_wmma_scale16_f32_16x16x128_f8f6f4 : AMDGPUWmmaScaleIntrinsicModsC<llvm_i64_ty>;
def int_amdgcn_wmma_f32_32x16x128_f4 : AMDGPUWmmaIntrinsicF4ModsC<llvm_anyint_ty, llvm_anyint_ty, llvm_anyfloat_ty>;
def int_amdgcn_wmma_scale_f32_32x16x128_f4 : AMDGPUWmmaScaleF4IntrinsicModsC<llvm_i32_ty>;
def int_amdgcn_wmma_scale16_f32_32x16x128_f4 : AMDGPUWmmaScaleF4IntrinsicModsC<llvm_i64_ty>;
}

class AMDGPUSWmmacIntrinsicABIdx<LLVMType A, LLVMType B, LLVMType CD, LLVMType Index> :
Expand Down
2 changes: 2 additions & 0 deletions llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4801,6 +4801,8 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
case Intrinsic::amdgcn_wmma_scale_f32_16x16x128_f8f6f4:
case Intrinsic::amdgcn_wmma_scale16_f32_16x16x128_f8f6f4:
case Intrinsic::amdgcn_wmma_f32_32x16x128_f4:
case Intrinsic::amdgcn_wmma_scale_f32_32x16x128_f4:
case Intrinsic::amdgcn_wmma_scale16_f32_32x16x128_f4:
case Intrinsic::amdgcn_swmmac_f16_16x16x64_f16:
case Intrinsic::amdgcn_swmmac_bf16_16x16x64_bf16:
case Intrinsic::amdgcn_swmmac_f32_16x16x64_bf16:
Expand Down
19 changes: 19 additions & 0 deletions llvm/lib/Target/AMDGPU/VOP3PInstructions.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1763,6 +1763,8 @@ def F16_FP8BF8X64_WMMA_w32 : VOP3PWMMA_Profile<[v8f16, v8i32, v8i32, v8f16
def F16_FP8BF8X128_WMMA_w32 : VOP3PWMMA_Profile<[v8f16, v16i32, v16i32, v8f16], 0, 0, 0, 1, 1, 0, 0, 0, 1>;
def F32_32X16X128_F4_WMMA_w32 : VOP3PWMMA_Profile<[v16f32, v16i32, v8i32, v16f32], 0, 0, 0, 0, 1, 0, 0, 0, 0, 1>;
def I32_IU8X64_WMMA_w32 : VOP3PWMMA_Profile<[v8i32, v8i32, v8i32, v8i32], 0, 0, 1, 0, 1, 0, 0, 0, 1>;
def F32_32X16X128_F4_SCALE_w32 : VOP3PWMMA_Profile<[v16f32, v16i32, v8i32, v16f32], 0, 0, 0, 1, 1, 0, 1, 0, 1>;
def F32_32X16X128_F4_SCALE16_w32 : VOP3PWMMA_Profile<[v16f32, v16i32, v8i32, v16f32], 0, 0, 0, 1, 1, 0, 1, 1, 1>;
def F32_F16X64_SWMMAC_w32 : VOP3PWMMA_Profile<[v8f32, v16f16, v32f16, v8f32], 1, 16, 0, 0, 1, 0, 0, 0, 1>;
def F32_BF16X64_SWMMAC_w32 : VOP3PWMMA_Profile<[v8f32, v16bf16, v32bf16, v8f32], 1, 16, 0, 0, 1, 0, 0, 0, 1>;
def F16_F16X64_SWMMAC_w32 : VOP3PWMMA_Profile<[v8f16, v16f16, v32f16, v8f16], 1, 16, 0, 0, 1, 0, 0, 0, 1>;
Expand Down Expand Up @@ -1852,6 +1854,9 @@ defm V_SWMMAC_F16_16X16X64_F16_w32 : SWMMACInstGFX12<"v_swmmac_f16_16x16x64
defm V_WMMA_F32_16X16X128_F8F6F4 : WMMAInst_SrcFormats_mc<"v_wmma_f32_16x16x128_f8f6f4", "F32_16X16X128_F8F6F4">;
defm V_WMMA_SCALE_F32_16X16X128_F8F6F4 : WMMAInst_SrcFormats_mc<"v_wmma_scale_f32_16x16x128_f8f6f4", "F32_16X16X128_F8F6F4_SCALE">;
defm V_WMMA_SCALE16_F32_16X16X128_F8F6F4 : WMMAInst_SrcFormats_mc<"v_wmma_scale16_f32_16x16x128_f8f6f4", "F32_16X16X128_F8F6F4_SCALE16">;

defm V_WMMA_SCALE_F32_32X16X128_F4_w32 : WMMAInstGFX12<"v_wmma_scale_f32_32x16x128_f4", F32_32X16X128_F4_SCALE_w32, "_w32">;
defm V_WMMA_SCALE16_F32_32X16X128_F4_w32 : WMMAInstGFX12<"v_wmma_scale16_f32_32x16x128_f4", F32_32X16X128_F4_SCALE16_w32, "_w32">;
} // End is_wmma_xdl = 1.

defm V_WMMA_LD_SCALE_PAIRED_B32 : VOP3PInst<"v_wmma_ld_scale_paired_b32", VOP_WMMA_LD_SCALE<i32, VCSrc_b32>>;
Expand Down Expand Up @@ -2010,6 +2015,8 @@ let SubtargetPredicate = isGFX125xOnly in {
defm : WMMAPat<"V_WMMA_F32_16X16X128_BF8_FP8_w32", int_amdgcn_wmma_f32_16x16x128_bf8_fp8, F32_FP8BF8X128_WMMA_w32>;
defm : WMMAPat<"V_WMMA_F32_16X16X128_BF8_BF8_w32", int_amdgcn_wmma_f32_16x16x128_bf8_bf8, F32_FP8BF8X128_WMMA_w32>;
defm : WMMAPat<"V_WMMA_F32_32X16X128_F4_w32", int_amdgcn_wmma_f32_32x16x128_f4, F32_32X16X128_F4_WMMA_w32>;
defm : WMMAPat<"V_WMMA_SCALE_F32_32X16X128_F4_w32", int_amdgcn_wmma_scale_f32_32x16x128_f4, F32_32X16X128_F4_SCALE_w32>;
defm : WMMAPat<"V_WMMA_SCALE16_F32_32X16X128_F4_w32", int_amdgcn_wmma_scale16_f32_32x16x128_f4, F32_32X16X128_F4_SCALE16_w32>;

foreach I = ["f8_f8", "f8_f6", "f8_f4", "f6_f8", "f6_f6", "f6_f4", "f4_f8", "f4_f6", "f4_f4"] in {
defm : WMMAPat<"V_WMMA_F32_16X16X128_F8F6F4_" # I # "_w32", int_amdgcn_wmma_f32_16x16x128_f8f6f4, !cast<VOP3PWMMA_Profile>("F32_16X16X128_F8F6F4_" # I # "_w32")>;
Expand Down Expand Up @@ -2191,6 +2198,15 @@ class VOP3PX2e <bits<8> op, bits<8> LdScaleOp, VOP3PWMMA_Profile P> : Enc128, VO
let Inst{127} = !if(P.NegLo2, src2_modifiers{0}, 0);
}

multiclass VOP3PX2_Real_ScaledWMMA_F4<bits<8> op, bits<8> LdScaleOp, VOP3PWMMA_Profile WMMAP> {
defvar PS = !cast<VOP3P_Pseudo>(NAME # "_twoaddr");
let SubtargetPredicate = isGFX1250Plus, WaveSizePredicate = isWave32,
DecoderNamespace = "GFX1250" in {
def _gfx1250 : VOP3P_Real_Gen<PS, GFX1250Gen, PS.Mnemonic>,
VOP3PX2e <op, LdScaleOp, WMMAP>;
}
}

multiclass VOP3PX2_Real_ScaledWMMA<bits<8> op, bits<8> LdScaleOp, VOP3PWMMA_Profile WMMAP> {
defvar PS = !cast<VOP3P_Pseudo>(NAME # "_twoaddr");
defvar asmName = !substr(PS.Mnemonic, 0, !sub(!size(PS.Mnemonic), !size("_f8_f8_w32")));
Expand Down Expand Up @@ -2292,6 +2308,9 @@ defm V_WMMA_F32_16X16X128_F8F6F4 : VOP3P_Real_WMMA_gfx1250_SrcFormats<0x0
defm V_WMMA_SCALE_F32_16X16X128_F8F6F4 : VOP3PX2_Real_ScaledWMMA_SrcFormats<0x033, 0x35, "F32_16X16X128_F8F6F4_SCALE">;
defm V_WMMA_SCALE16_F32_16X16X128_F8F6F4 : VOP3PX2_Real_ScaledWMMA_SrcFormats<0x033, 0x3a, "F32_16X16X128_F8F6F4_SCALE16">;

defm V_WMMA_SCALE_F32_32X16X128_F4_w32 : VOP3PX2_Real_ScaledWMMA_F4<0x088, 0x35, F32_32X16X128_F4_SCALE_w32>;
defm V_WMMA_SCALE16_F32_32X16X128_F4_w32 : VOP3PX2_Real_ScaledWMMA_F4<0x088, 0x3a, F32_32X16X128_F4_SCALE16_w32>;

defm V_SWMMAC_F32_16X16X64_F16_w32 : VOP3P_Real_WMMA_gfx1250 <0x065, F32_F16X64_SWMMAC_w32>;
defm V_SWMMAC_F32_16X16X64_BF16_w32 : VOP3P_Real_WMMA_gfx1250 <0x066, F32_BF16X64_SWMMAC_w32>;
defm V_SWMMAC_F16_16X16X64_F16_w32 : VOP3P_Real_WMMA_gfx1250 <0x067, F16_F16X64_SWMMAC_w32>;
Expand Down
Loading